Financial Securities Based On Neural Network Prediction Method

The financial securities market is most complex nonlinear system, the changing of its time series involves many uncertainty factors such as politics, economic, psychology, meanwhile the interaction among all the factors is nonlinear and time changing, having inherence noisy, random, nonstationary while self-organized, self-adjusted, periodical, tendency and approaching a certain rule. The interaction relationship between the input and output of financial time series will gradual change time by time, and traditional only using linear analytic methods have localization and can hardly forecast accurate results. Stand or fall of the securities forecast has great significance for a country's economic development and hugeness investors. The purpose of the research in this paper is to combine fuzzy neural network and support vector machine, two of the most important nonlinear analytic methods, with financial securities theory, according to the most complexity nonlinear characteristics of the information of financial securities market, study and forecast the information of the financial securities though basing the research on a more scientific, precise, intelligent theoretic foundation by constructing models, extracting features, and analyzing the time series data of the financial securities market.This paper investigates stock prediction of financial securities market and studies models of fuzzy neural networks and support vector machine deeply into the foresting theories and actual results of time series of China financial securities markets, analyses, compares and evaluates on the results of the research. The main work of this paper is following:1. The paper uses the new low complex fuzzy active function into the design of the structure of fuzzy neural network and algorithm in stock prediction, realizes the self-adaptive of member function and the self-organization of the fuzzy rules by its self-learning and competition. The simulation experimentation shows that this scheme makes the algorithm simpler, convergence, computation low complexity, tend to the explanation of the if-then rules and the realization of hardware. The method of the prediction has the higher precision, distinct affective and superiority.2. The paper considers the nonlinear features of the financial securities market and the psychology of the investors and enhancing the precision of the prediction and the capacity of generalization, proposes the self-adaptive reduced support vector machine (RASVM) by combining the self-adaptive parameters SVM into the reduced support vector machine (RSVM). The experiment shows that the performance of the prediction of RASVM surpasses the standard SVM. RASVM possesses less convergence support vectors, training time and higher precision of prediction, capacity of generalization. RASVM can obtain higher performance than the standard SVM in the stock prediction. The study methods this paper proposes can also be applied in others financial securities time series forecasting.